With the growth of population and the rapid development of society, the problem of water shortage and water pollution have become increasingly serious. As an important aspect of water pollution assessment, water quality classification has a more prominent significance and role. Based on terahertz attenuated total reflection (THz-ATR) spectrum and pattern recognition technology, a water quality analysis model is proposed in this paper. The terahertz time-domain spectroscopy system and the attenuated total reflection module were used to measure the terahertz attenuated total reflection spectra of five water samples, including pure water, tap water, river water, seawater A and seawater B. The refractive index, absorption coefficient, real and imaginary parts of the dielectric constant of five water samples in the frequency range of 0.2~1.0 THz were obtained using the optical parameter extraction model. Principal Component Analysis (PCA) was applied to conduct refractive index reduction and feature extraction, and two-dimensional score charts of the first and second principal components and three-dimensional score charts of the first three principal components of the samples were made respectively. It can be seen that the principal component score chart based on the index of refraction can clearly distinguish different water samples. In order to further classify different water samples accurately, the data after dimension reduction is input into a support vector machine to construct a water sample classification model. Three-fifths of each water sample is randomly selected as the training set, and the remaining data is used as the test set. At the same time, three optimization algorithms, grid search (GridSearch), genetic algorithm (GA) and particle swarm algorithm (PSO) are introduced to optimize the parameters of the support vector machine. The results show that the optimal parameters c and g of the support vector machine based on the grid search algorithm are 1.414 2 and 2.0, respectively, with an accuracy of 99.0%; the optimal parameters c and g of the support vector machine based on the genetic algorithm are 1.675 4 and 5.966 5, respectively, which are accurate The rate is 99.5%; the optimal parameters c and g of the support vector machine based on particle swarm optimization are 3.154 9 and 12.589 respectively, and the accuracy rate is 100%. It can be seen that the optimal parameters obtained by different optimization algorithms are different, and all the SVM classification models constructed can achieve correct classification, and the classification accuracy is above 99.0%. The research results show that the PCA-SVM classification model based on the refractive index constructed by the particle swarm optimization algorithm has the best effect and can accurately identify different water samples, laying a foundation for water quality classification.